AI Policy · Daily

BIS export license backlog is blocking billions in AI chip sales to US allies, directly undermining the Trump administration's push to use chip access as a geopolitical tool. One day after Treasury and Fed officials convened an emergency cybersecurity summit, Wall Street banks began Mythos testing at government urging. PJM Interconnection filed an emergency proposal with FERC to add 15 gigawatts of capacity across 13 states to meet surging AI data center demand. The Musk trial begins April 27, and OpenAI separately faces a stalking victim's lawsuit over an ignored mass-casualty warning. The FCC proposed a Chinese telecom data center ban that would go further than any FCC action since it revoked China Telecom Americas' authorization in 2022.

I.AI Policy Today

BIS staffing attrition stalls billions in AI chip export licenses to US allies

The Bureau of Industry and Security faces licensing backlogs, staff departures, and unclear policy direction that are blocking the Trump administration's push to expand AI chip sales to allied nations, per Bloomberg. The agency responsible for processing sensitive technology export licenses is losing experienced staff faster than it can replace them, creating a bottleneck that has stalled billions of dollars in pending transactions. The dysfunction directly undermines a presidential directive to accelerate chip exports to partners as a geopolitical tool. Industry officials told Bloomberg they have no visibility into when the backlog will clear.

Read at Bloomberg ↗

Wall Street banks begin internal testing of Anthropic's Mythos model at government urging

Major banks are deploying Anthropic's Mythos model internally for vulnerability detection, one day after Treasury Secretary Bessent and Fed Chair Powell convened an emergency summit warning of AI-driven cybersecurity threats, per Bloomberg. Trump administration officials are actively encouraging the financial sector to adopt the model, which found working exploits for zero-day vulnerabilities 72.4% of the time during Anthropic's controlled deployment earlier this week. Which specific banks are testing and at what scale has not been disclosed.

Read at Bloomberg ↗

PJM Interconnection files emergency proposal with FERC for 15 gigawatts of new data center power

The nation's largest regional grid operator, serving 65 million people across 13 states and Washington, D.C., is seeking emergency approval from the Federal Energy Regulatory Commission to add 15 gigawatts of generating capacity to meet surging AI data center demand, per Bloomberg. Fifteen gigawatts is roughly equivalent to the electricity consumption of 10 million homes. PJM's service territory includes Virginia's data center corridor, the largest concentration of data centers in the world. FERC review is expected to begin immediately.

Read at Bloomberg ↗

OpenAI accuses Musk of "legal ambush" as $100 billion-plus trial approaches

OpenAI said in court filings that Elon Musk abruptly shifted his legal strategy weeks before trial in a case involving more than $100 billion in potential liability, per Bloomberg. OpenAI characterized the move as a deliberate ambush; Musk's representatives have not publicly responded to the characterization. Musk and OpenAI are direct competitors in the frontier AI market through xAI and ChatGPT, respectively. Jury selection is scheduled to begin April 27 in Oakland.

Read at Bloomberg ↗

Stalking victim sues OpenAI, alleges company ignored three warnings including its own mass-casualty flag

A stalking victim filed suit against OpenAI on Friday, alleging the company ignored three separate warnings that a ChatGPT user posed a threat to her safety, including an internal system flag indicating potential mass-casualty risk, per TechCrunch. The plaintiff's attorneys at Edelson PC, the same firm behind the Adam Raine teen suicide case, filed a temporary restraining order requesting that OpenAI block the user, preserve chat logs, and notify the plaintiff of new account creation attempts. OpenAI agreed to suspend the user's account but declined the other measures. The lawsuit was filed the same week OpenAI testified in favor of an Illinois bill that would shield AI companies from liability below a 100-death threshold.

Read at TechCrunch ↗

II.China Watch

FCC proposes barring Chinese telecoms from operating US data centers

The Federal Communications Commission said Thursday it is considering prohibiting China Mobile, China Telecom, and China Unicom from operating data centers on US soil and banning American carriers from interconnecting with them, per SCMP citing Reuters. The proposal could extend to affiliates using Huawei or ZTE equipment and would go further than any FCC telecom restriction since it revoked China Telecom Americas' authorization in 2022. An initial vote is scheduled for the FCC's April 30 meeting.

Read at SCMP ↗

DeepSeek recruits data center engineers for facilities reportedly using banned Nvidia chips

DeepSeek is hiring data center engineers for facilities in Inner Mongolia where it reportedly relies on Nvidia Blackwell chips barred from export to China, per Bloomberg. A Trump administration official alleged in February that DeepSeek trained its V4 model on thousands of smuggled Blackwell GPUs; Nvidia has called the claim "farfetched." BIS has not announced new enforcement actions related to the allegations.

Read at Bloomberg ↗

Shengshu Technology raises $290 million for "world model" AI bridging video and robotics

Beijing-based Shengshu Technology completed a nearly 2 billion yuan Series B round led by Alibaba Cloud, with participation from the state-backed China Internet Investment Fund, per CNBC. The startup behind the Vidu video generation platform is building "universal world models" designed to bridge digital content generation and physical robotics control. Shengshu's Vidu platform currently generates video content and is being adapted for robotics applications.

Read at CNBC ↗

III.Federal Policy Tracker

President Trump publicly praises Palantir's "great war fighting capabilities" during Iran conflict

President Trump posted on Truth Social praising Palantir's military AI capabilities by name while the company's Maven platform is reportedly deployed in US operations against Iran, per CNBC. The post briefly boosted Palantir's stock price before broader market pressure reasserted itself. Palantir stock ended the week with its steepest decline in over a year despite the presidential endorsement.

Read at CNBC ↗

IV.Industry & Market Watch

Anthropic signs multibillion-dollar deal with CoreWeave to power Claude

Anthropic agreed to rent CoreWeave's AI computing capacity in a multiyear, multibillion-dollar arrangement to support development and deployment of Claude, per Bloomberg. CoreWeave will provision a variety of Nvidia chips in US data centers for the deal. CoreWeave now supplies infrastructure to both Anthropic and Meta, which signed its own $21 billion CoreWeave expansion earlier this week; CoreWeave shares rose 11% on the Anthropic announcement.

Read at Bloomberg ↗

Google and Intel expand AI infrastructure partnership across multiple chip generations

Google Cloud will adopt multiple future Intel Xeon processor iterations and the companies will co-develop custom ASIC-based infrastructure processing units, per TechCrunch. The deal arrives as cloud providers face a growing CPU shortage driven by AI infrastructure scaling. Intel shares rose 4.7% on the announcement; Intel CEO Lip-Bu Tan characterized the partnership as addressing the need for balanced AI systems across multiple generations of silicon.

Read at TechCrunch ↗

RISC-V chip design firm SiFive raises $400 million at $3.65 billion valuation

SiFive closed a $400 million Series G led by Atreides Management, with participation from Nvidia and Apollo, per SiliconANGLE. The investment values SiFive at $3.65 billion and brings total outside funding above $760 million. SiFive's open-source RISC-V processor designs are used in more than 500 chip projects by major technology firms; RISC-V is not subject to the same licensing restrictions as proprietary x86 and Arm architectures.

Read at SiliconANGLE ↗

Cisco reportedly in talks for second AI security acquisition in two days

Cisco is in discussions to acquire AI agent security startup Astrix Security for $250 to $350 million, per SiliconANGLE citing The Information. The deal would follow Cisco's purchase of AI observability firm Galileo Technologies one day earlier. Astrix helps companies discover, identify, and secure AI agents operating within their enterprise networks; the company was backed by Workday and Anthropic's Anthology Fund.

Read at SiliconANGLE ↗

V.Global & Geopolitics

Bank of Canada convenes major lenders over Anthropic Mythos cybersecurity risks

The Bank of Canada and major Canadian financial institutions met Friday to discuss cybersecurity threats posed by Anthropic's Mythos model, per Bloomberg. The meeting followed the emergency summit convened the day before in Washington by Treasury Secretary Bessent and Fed Chair Powell, with both governments responding to the same vulnerability findings. No public statement has been issued on the outcomes of the Canadian meeting.

Read at Bloomberg ↗